Analyzing the Future

Every business would like to have an accurate crystal ball. But since they don't exist, predictive analytics is the next best thing.

Predictive analytics is the systematic study of historical results, of sales or other aspects of a business, to determine patterns and predict future performance. It's almost always done with iterative software that takes input, often supplied continuously, and usually makes preprogrammed adjustments for seasonal, cyclical and other factors.

Predictive analytics may be systematic, but there is nothing automatic or easy about designing or using it. Identifying and prioritizing the factors that can affect performance–whatever type of performance that may be–and deciding how much weight to give them is one of the toughest jobs in business planning.

For this reason, predictive analytics has made more progress in some aspects of CPG retailing than in others. It all depends on the "organizational culture," says Anbu Mani, advisory partner in retail and consumer practice at PwC.

"More often than not, organizations have gone primarily from [being] institution driven–meaning, 'I've been here for many years, so trust me, it will work'–to more data-driven," Mani says. "Decisions are made with facts and data that support the decision-making process."

SHOPPER OUTREACH

Broadly speaking, the most widespread, developed use of predictive analytics in CPG retailing, say industry observers, is in marketing and shopper outreach.

"That's where we have seen the marketing function as a whole leverage predictive analytics to get close to consumers, be it a loyal customer or a one-off customer," Mani says. "With either scenario, there is a big push to understand who the customer is and how can we address their needs to provide a personalized experience across all channels."

Personalization, or "marketing to the customer," is accomplished by applying predictive analytics to as narrow a group of shoppers as possible.

"Customer loyalty analysis is used to optimize customer segmentation to improve upsell and cross-sell opportunities," says Russ Hill, senior director of global retail industry marketing and analytics product marketing for SAP. "This is managed through stronger relationship definition in order to create customer segmentations, identify the most loyal and profitable customers, and develop targeted promotions."

The idea is to narrow the target a much as possible, says Jed Alpert, vice president of marketing for consulting firm 1010data.

"CPGs and retailers have used many modeling techniques to cluster consumers into 'like groups' as well as predict their shopping patterns," Alpert says. "Typically these are on an aggregated/sampled data set and not on a per consumer basis, reducing their accuracy."

Obviously, the most desirably "narrow" target would be shoppers as individuals. That's within the reach, but not yet the grasp, of retailers, thanks to ever-increasing Big Data, Alpert says: "The accumulation of large amounts of data (Big Data) on each shopper is opening the door to building much better models of shopper behavior."

The primary mechanism for shopper personalization is the loyalty program, which is, at its core, a record of every shopper's purchases. This forms a basis on which to build individual offers. But it imposes a structure on the store's pricing and other operations that some retailers find constraining. Some, like Albertsons last year, have moved away from loyalty programs.

"It's important to note that there are two types of food retailers: those that have a loyalty program and believe in its value, and those that are adamantly against 'tiering' customers," Hill says, noting that Kroger is enthusiastically expanding its loyalty program. "The questions begged of this 'evolution of customer loyalty' is whether this is a cost-cutting move, and how the company will leverage data in the new structure, in the case of Albertsons. For Kroger and others, it's a matter of ramping up the use of customer data for improving the customer experience."

Whether a retailer uses a loyalty program or not, the basics of shopper outreach using predictive analytics remain the same, Hill says: It's about accurately determining the behavior of a given segment of shoppers, and comparing that behavior to those of shoppers as a whole. That provides guidance for "stretch" offers, for more of an item than a shopper usually buys; cross-selling of items that are perceived to go with what a shopper usually buys; upselling; targeting of print and digital offers, and other aspects of shopper marketing.

INVENTORY MANAGEMENT

Another important aspect of predictive analytics is in ordering and maintaining inventory. It can help with minimizing out-of-stocks, preparing for sales cycles and seasonal events, planning and managing promotions and discounts, and other aspects of inventory management.

Mani of PwC notes that while many CPG retailers use predictive analytics to avoid stockouts in individual stores, it's being applied inconsistently across brands and SKUs.

"That's an area where we see, from our vantage point, companies trying to make a push in terms of aligning the inventory, planning and executing aspects of the supply chain and understanding needs at the zip code level," he says.

Information that can be gleaned from the use of predictive analytics, and action that can be taken based on that information, includes store clustering, the effects of sales cannibalization and other factors caused by multiple outlets in one geographic area; out of stock frequency, where the stores and items that run most often out of stock can be identified and when it's most likely to happen; forecasting demand and stocks necessary for new products; and trade promotion optimization.

Arranging trade promotions for maximum impact is one of the trickiest aspects of high-volume CPG retailing, and it's being underutilized, says Kurt Jetta, CEO of TABS Group, a business consultancy that specializes in predictive analytics. Neither predictive analytics nor the concept of trade promotion itself are being used effectively by most retailers, Jetta says.

"If there is a silver bullet in this whole panorama of CPGs, that is it," Jetta says. "I would contend [predictive analytics is] not being often or effectively used generally. If companies were adequately addressing and taking advantage of predictive analytic opportunities, you would certainly expect those numbers [CPG retail growth] to be more robust."

A good trade optimization program can identify the degree of price elasticity in a given product or even SKU, and which price points diminish or eliminate that elasticity. It can help settle on a profitable baseline price, determine the effects of displays and other special merchandising, deal with "outliers" or statistical flukes, and suggest the timing and other tactics best suited to promotions for a given retailer.

PEOPLE PREDICTIONS

Another area where predictive analytics have even more unutilized potential is in human resources, Mani says. In retailing, where personal interactions are often important, it's vital to have the right talent–and predictive analytics can help find and retain it.

"You as a retailer should have a way to identify, attract and retain the right set of talent over a period of time," Mani says. "And I see how predictive analytics could play a significant role in that space." He says such a system could identify the best colleges for graduating talent and keep track of its graduates who are hired, measuring their performance in different positions.

Given all these possibilities, it's perhaps not surprising that the first task of a retailer looking to enhance predictive analytics is to decide exactly what it's supposed to do.

"Whoever is doing this on the analytical side [needs] to really have a very good knowledge of what are the business issues most important for retailers," says Eugene Roytburg, managing partner of 4i, a growth and foresight analytics firm. "Existing and new business issues that haven't been addressed by predictive analytics yet....People have been buying different tools and processes, developing those functions, without clearly understanding the value that analytics can create."

THE HUMAN FACTOR

Once it's up and running in the real world, of course, further adjustments often must be made. That usually requires the human touch.

"It is common practice to monitor the level of error (or success) that a model has generated and then adjust as needed," says Alpert of 1010data. "Automating the detection of the error is rather simple, where adjusting the error is not so simple and will require an analyst (statistician)."

As with any software, predictive analytics applications are only as good as the data they receive. Retailers often focus on the accuracy of the predictive model and not the accuracy or completeness of the data fed into it, Mani says.

"Human business acumen, plus the model coming together, is where you get the maximum results," Mani says. He recalled a client who asked PwC with help with chronic stockouts that its analytical software kept failing to predict. Once PwC looked into the situation, it became clear that the software had no way of addressing key business events or tracking certain vital SKUs. PwC gathered information from store personnel and revised the model to take these events into account, accuracy greatly improved.

One of the challenges with predictive analytics is for retailers to think of it as a function of a specific part of the organization, like supply chain or marketing, rather than a bunch of results that the IT people generate and then hand off to other departments.

"More often than not, [predictive analytics is] discussed in terms of a technology capability, with the technology teams driving it," Mani says. "But as you walk through the continuum, more and more businesses are realizing that there's value to keep them as a function that drives analytics services across all parts of the organization."